Class CLIPCapDataset

utils/clipscore/clipscore.py:66–79

class CLIPCapDataset(torch.utils.data.Dataset):
    def __init__(self, data, prefix='A photo depicts'):
        self.data = data
        self.prefix = prefix
        if self.prefix[-1] != ' ':
            self.prefix += ' '

    def __getitem__(self, idx):
        c_data = self.data[idx]
        c_data = clip.tokenize(self.prefix + c_data, truncate=True).squeeze()
        return {'caption': c_data}

    def __len__(self):
        return len(self.data)


class CLIPImageDataset(torch.utils.data.Dataset):